Method and system for detecting wear particles in lubricating oil

Abstract

The application discloses a kind of lubricating oil abrasive particle detection method and system, method includes data acquisition: arranging capacitor sensor on the outer wall of pipeline, and the initial capacitance data C1 of the region to be measured is obtained by capacitor sensor;Data processing one: the initial capacitance data C1 is processed by dimension reduction by principal component analysis method, and second-order capacitance data C2 is obtained;Data processing two: second-order capacitance data C2 is input into neural network model, constraint condition is added, and third-order capacitance data C3 is obtained;Image reconstruction: third-order capacitance data C3 is input into finite element analysis model, and the image in the region to be measured is obtained;Evaluation optimization: the reconstructed image is evaluated, if imaging is not clear, then repeat data processing two and image reconstruction process, until imaging quality meets requirements.In the application, data is analyzed by principal component analysis method, which helps to establish a more simplified neural network model, the entire process consumes less time, the amount of calculation is moderate, and the accuracy can be guaranteed.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of testing materials by electromagnetic technology, and in particular to a method and system for detecting particles in lubricating oil. BACKGROUND

[0002] Lubricating oil is a commonly used medium in the industrial field, especially in some transmission fields. It mainly plays a role in full lubrication and temperature reduction.

[0003] During the use of lubricating oil, some particulate impurities inevitably exist, which may affect the normal use performance of the lubricating oil. Therefore, in some related technologies, it is usually necessary to monitor the lubricating oil to know the situation of the particles inside the lubricating oil, so as to take corresponding correct measures. Especially in some use fields with safety requirements, such as auxiliary power units (APU), the particle situation in the lubricating oil is an important factor affecting flight safety.

[0004] Currently, there are offline detection or online detection methods, such as sampling detection in offline detection, but offline detection cannot obtain the lubricating oil in real time, and the immediacy is poor and the reliability is insufficient; online detection has a capacitive tomography technology, which can image the medium in the pipeline in real time, but the current online detection still has the defects of large calculation amount and insufficient accuracy.

[0005] SUMMARY

[0006] The present application provides a method and system for detecting particles in lubricating oil, in order to solve the technical problems of large calculation amount and insufficient accuracy in the current online detection.

[0007] In a first aspect, the present application provides a method for detecting particles in lubricating oil, comprising: data acquisition: arranging a capacitive sensor on the outer wall of the pipeline, and obtaining initial capacitance data C1 of the to-be-detected region by the capacitive sensor; data processing I: performing dimensionality reduction processing on the initial capacitance data C1 by a principal component analysis method to obtain second-order capacitance data C2; data processing II: inputting the second-order capacitance data C2 into a neural network model, adding a constraint condition, and obtaining third-order capacitance data C3; image reconstruction: inputting the third-order capacitance data C3 into a finite element analysis model to obtain an image in the to-be-detected region; evaluation and optimization: evaluating the reconstructed image, and if the imaging is not clear, repeating the data processing II and image reconstruction process until the imaging quality meets the requirements.

[0008] In the embodiment of the present application, the dimensionality reduction processing of the initial capacitance data C1 by the principal component analysis method can effectively reduce the dimensionality of the input data, remove redundant information and only retain the most important feature set in the initial capacitance data C1. Considering that the principal component analysis method performs linear dimensionality reduction, there are some nonlinear original data in the initial capacitance data C1. Therefore, only using the principal component analysis method may not be able to capture the important features of the initial capacitance data C1. Therefore, the second-order capacitance data C2 after dimensionality reduction is further processed, that is, processed by a neural network model. The neural network model optimizes the algorithm according to the second-order data C2 under the action of the constraint condition and outputs the third-order capacitance data C3. Since the neural network model can capture, fit and other operations on nonlinear features, a suitable neural network model can be selected or designed in the actual operation process, so that the neural network model can establish a relatively accurate mapping relationship between the second-order capacitance data C2 and the third-order capacitance data C3. Therefore, the neural network model can capture or improve the data missed by the principal component analysis method. In this way, the second-order capacitance data C2 after the principal component analysis reduces the computational complexity of the neural network model, speeds up the training speed, reduces the risk of overfitting, and helps the neural network model to better learn the key features in the data, thereby establishing a more concise and accurate neural network model. And the neural network model can perform secondary processing on the second-order capacitance data C2 to finally output accurate third-order capacitance data C3. The entire process consumes less time, has moderate computational complexity, and can ensure data accuracy. Finally, the finite element analysis model is used for image reconstruction. The third-order capacitance data C3 is data after dimensionality reduction and neural model improvement, which can maintain a certain accuracy and simplicity. After image reconstruction by the finite element analysis model, the time consumption is shorter and the result is more accurate.

[0009] As one of the optional embodiments of the present application, a step of establishing a neural network model is further included between data processing I and data processing II: forward problem solving: calculate the theoretical capacitance value distribution according to the collected initial capacitance data C1 and the characteristic parameters of the collection system, denoted as target capacitance data C0; set a neural network model suitable for processing capacitance data, select a suitable network structure and layer number, and add a constraint condition; take the second-order capacitance data C2 as input data and the target capacitance data C0 as output data, and train the neural network model through the second-order capacitance data C2 and the target capacitance data C0, so that the neural network model learns the mapping relationship between the input and output data.

[0010] In the embodiment of the present application, the target capacitance data C0 is obtained by solving the forward problem. The target capacitance data C0 is a set of data that can most accurately reflect the image of the to-be-measured region in theory, which is obtained by a series of rigorous formula derivation and calculation, and is very mature in the prior art, which will not be repeated here. The establishment process of the traditional neural network model is as follows: input the initial capacitance data C1 and the target capacitance data C0, and train multiple times to let the neural network learn the mapping relationship between C1 and C0, and then input the corresponding initial capacitance data C1 to obtain the capacitance data used for image reconstruction. However, there are obvious problems, that is, the initial capacitance data C1 has a large amount of data interference, and the data volume itself is large, which greatly increases the learning time and load degree of the neural network model, and also greatly interferes with the mapping accuracy of the neural network model, which greatly hinders the instantaneity of image reconstruction. In the present application, the initial capacitance data C1 is processed by the principal component analysis method, and the processed second-order capacitance data C2 is input during the training of the neural network. Then the neural network model only needs to learn the relationship between C2 and C0. In this way, the training process of the neural network model is greatly reduced, and since the guide capacitance data is also the target capacitance data C0, the third-order capacitance data C3 used for the image reconstruction process can be obtained when C2 is input subsequently, and the third-order capacitance data C3 is relatively close to the target capacitance data C0.

[0011] As one of the optional embodiments of the present application, the steps of solving the forward problem include: establishing a finite element model: dividing the target object into discrete small unit cells, each unit cell having a certain electrical conductivity; constructing a forward problem equation: constructing a forward problem equation according to the initial capacitance data C1 and the finite element model, the forward problem equation being used to indicate the relationship between the initial capacitance data C1 and the electrical conductivity of the to-be-measured region; solving the forward problem: obtaining the electrical conductivity distribution of the to-be-measured region by solving the forward problem equation, denoted as the target capacitance data C0.

[0012] As one of the optional embodiments of the present application, the step of evaluating and optimizing further includes: if the results of multiple image reconstructions do not meet the expectations, replacing different types of neural network models and repeating the step of establishing the neural network model, wherein the neural network model includes a convolutional neural network, a recurrent neural network and a feedforward neural network.

[0013] In the embodiment of the present application, there are many types of neural network models, and the adaptability is also different. If the results of multiple image reconstructions do not meet the expectations, it proves that the neural network model may have the problem of poor adaptability. In the process of evaluating and optimizing, the neural network model is changed to find the most suitable neural network model, thereby improving the accuracy of subsequent data processing.

[0014] As one of the optional embodiments of the present application, the step of evaluating optimization further comprises: if the results of multiple image reconstruction do not meet the expectations, changing the combination of constraint conditions for repeated training, wherein the constraint conditions include physical constraints, spatial constraints, prior knowledge constraints, etc.

[0015] In the embodiments of the present application, the combination of constraint conditions is changed multiple times, and the constraint condition most suitable for the current detection process can be found in the neural network training process, so as to remove the noise in the capacitance data to the greatest extent, improve the smoothness of the data, and further improve the spatial resolution of the capacitance image, so that the imaging result is more accurate and detailed.

[0016] As one of the optional embodiments of the present application, the step of data processing further comprises a step of data preprocessing: processing the abnormal values and noise in the initial capacitance data C1.

[0017] In the embodiments of the present application, in the principal component analysis method, there may be a case of insufficient data capture, which is due to the collection of abnormal values or noise in the initial capacitance data C1 caused by circuit noise, installation position of the capacitance sensor, and precision control, etc. Since the principal component analysis method is based on the calculation of the covariance matrix or the correlation matrix of the data, the existence of abnormal values or noise may cause the calculation result of the covariance matrix to deviate from the true situation, thereby affecting the extraction and interpretation of the principal components. Therefore, the data preprocessing in advance can reduce the influence on the result of the principal component analysis process and ensure the accuracy of the result.

[0018] As one of the optional embodiments of the present application, the step of performing dimensionality reduction processing on the initial capacitance data C1 by the principal component analysis method to obtain the second-order capacitance data C2 comprises: performing standardization processing on the initial capacitance data C1, so that the mean value of each feature is 0 and the variance is 1; calculating the covariance matrix of the standardized capacitance data C1; performing eigenvalue decomposition on the covariance matrix to obtain the eigenvalues and the corresponding eigenvectors; sorting the eigenvalues in descending order, and selecting the eigenvectors corresponding to the first k eigenvalues as the principal components; projecting the standardized capacitance data C1 onto the selected principal components to obtain the dimensionality-reduced second-order capacitance data C2.

[0019] As one of the optional embodiments of the present application, the characteristic parameters of the acquisition system include: the geometric shape of the capacitance sensor, the electrode spacing, the electrode size, etc.

[0020] As one of the optional embodiments of the present application, the step of evaluating optimization further comprises: if the particle diameter D shown in the image is greater than or equal to 0.5 cm, or the image shows that the solubility is continuously high within each monitoring period T minutes, an alarm information is sent, wherein 15≤T≤25.

[0021] In another aspect, the application provides a lubricating oil detection system, comprising: a plurality of capacitive sensors arranged in a ring shape on the outer wall of a pipeline; a data acquisition module comprising a driving unit and a transmission unit, the driving unit being configured to apply an alternating voltage to stimulate a region to be measured; and a terminal configured to analyze and process the acquired capacitive data and reconstruct an image; wherein the transmission unit is electrically connected to the capacitive sensors and configured to convert the current signal into a digital signal and transmit the digital signal to the terminal.

[0022] The one of the above technical solutions has the following advantages or beneficial effects: the dimension reduction processing of the initial capacitive data C1 by the principal component analysis method can effectively reduce the dimension of the input data, remove redundant information, and only retain the most important feature set in the initial capacitive data C1. Considering that the principal component analysis method performs linear dimension reduction, there are some nonlinear original data in the initial capacitive data C1. Therefore, only using the principal component analysis method may not be able to capture the important features of the initial capacitive data C1. Therefore, the second-order capacitive data C2 after dimension reduction is further processed, that is, processed by a neural network model. The neural network model optimizes the algorithm according to the second-order data C2 under the action of the constraint condition and outputs the third-order capacitive data C3. Since the neural network model can capture, fit, etc. nonlinear features, a suitable neural network model can be selected or designed in actual operation to enable the neural network model to establish a relatively accurate mapping relationship between the second-order capacitive data C2 and the third-order capacitive data C3. Therefore, the neural network model can capture or improve the data missed by the principal component analysis method. In this way, the second-order capacitive data C2 after the principal component analysis reduces the computational complexity of the neural network model, speeds up the training, reduces the risk of overfitting, and helps the neural network model to better learn the key features in the data, thereby establishing a more concise and accurate neural network model. And the neural network model can perform secondary processing on the second-order capacitive data C2 to finally output accurate third-order capacitive data C3. The entire process consumes less time, has moderate computational complexity, and can ensure data accuracy. Finally, the finite element analysis model is used for image reconstruction. The third-order capacitive data C3 is data that has been dimensionally reduced and improved by the neural model. It can maintain a certain accuracy and conciseness. After image reconstruction by the finite element analysis model, the time consumed is shorter and the result is more accurate. BRIEF DESCRIPTION OF DRAWINGS

[0023] The technical solutions and other beneficial effects of the application will become apparent from the following detailed description of specific embodiments of the application, taken in conjunction with the accompanying drawings.

[0024] Fig. 1 is the overall flowchart of the lubricating oil abrasive particle detection method provided by the embodiments of the application.

[0025] Fig. 2 is a flowchart for establishing a neural network model provided by an embodiment of the present application;

[0026] Fig. 3 is a flowchart of a judgment logic after evaluation provided by an embodiment of the present application. DETAILED DESCRIPTION

[0027] The technical solutions in the embodiments of the present application will be clearly and completely described in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by a person skilled in the art without creative work fall within the protection scope of the present application.

[0028] In the description of the present application, it should be noted that, unless otherwise explicitly specified and limited, the term “and / or” in the present application is only to describe the association relationship of the associated objects, which means that there can be three relationships, for example, A and / or B can mean that there are three cases of A alone, A and B together, and B alone. In addition, the character “ / ” in the present application generally represents an “or” relationship between the associated objects without special description.

[0029] Related background introduction.

[0030] Electrical Capacitance Tomography (ECT) is a non-invasive imaging technology that can be used to feedback the internal fluid situation of an object. The imaging principle is that each phase of the measured substance has a different dielectric constant, and when the phase composition distribution or concentration distribution changes, the equivalent dielectric constant of the mixed fluid will change, thereby causing the capacitance value between the measurement electrode pairs to change.

[0031] ECT technology places multiple electrodes around an object to form a capacitive sensor array that surrounds the object. These electrodes record the amount of capacitance change in the object by alternately applying voltage and measuring current. In the imaging process, a small alternating voltage, called excitation voltage, is first applied around the region to be imaged. Then, a voltage is applied through the driving electrode to make the current flow through the object, and a pair of measurements is made between each electrode pair. When the current passes through the object, the differences in electrical properties (such as electrical conductivity) inside the object will cause the distribution of the current to change, thereby changing the capacitance value. By measuring multiple times, a series of capacitance data is obtained, which is sent to a computer for processing and reconstruction.

[0032] During the reconstruction process, the measured capacitance data is analyzed and processed by imaging algorithms to generate images. These algorithms include finite element method, gradient method, back projection method, etc. Among them, several commonly used image reconstruction algorithms are mainly introduced:

[0033] Finite Element Method (FEM): Finite Element Method is a widely used reconstruction algorithm in capacitance tomography. It divides the region to be imaged into many small finite element units, and models and calculates the capacitance change of each unit. By solving the positive problem of capacitance change, the measured data is fitted with the simulated data, so as to realize image reconstruction. Finite Element Method can consider complex capacitance distribution and object shape, but the calculation complexity is high.

[0034] Gradient Method: Gradient Method is a reconstruction algorithm based on mathematical gradient calculation. It calculates and optimizes the gradient of capacitance data to estimate the internal capacitance distribution of the object. Gradient method is usually an iterative algorithm, which gradually improves the reconstruction result through continuous iteration and optimization process. Gradient method has fast calculation speed and low complexity, but it has certain robustness problem for weak boundary and noise.

[0035] Back Projection Method: Back Projection Method is a simple and commonly used reconstruction algorithm. It is based on the projection principle, which projects the measured capacitance data back to the region to be imaged, so as to obtain the image. Back projection method has certain advantages in speed and calculation complexity, but it is sensitive to noise and nonlinearity.

[0036] The imaging principle of ECT technology is based on the electrical difference of the object, which can be widely used in the imaging of fluid distribution, solid particle distribution, fluid-solid interface positioning, etc. However, the imaging accuracy of ECT technology is affected by factors such as object shape, electrode arrangement, environmental noise, etc., resulting in insufficient clarity and accuracy of the current image reconstruction result.

[0037] The present application optimizes the image reconstruction process to improve the clarity and accuracy of imaging.

[0038] The following will be described in detail with reference to the accompanying drawings Figs. 1-3 Further description of the present application is made.

[0039] In a first aspect, the present application provides a method for detecting oil abrasive particles, comprising:

[0040] Step 1, data acquisition: arranging capacitance sensors on the outer wall of the pipeline, and acquiring initial capacitance data C1 of the region to be measured by the capacitance sensors;

[0041] Step2, data processing one: the initial capacitance data C1 is processed by principal component analysis method to reduce dimension, and second-order capacitance data C2 is obtained;

[0042] Step3, data processing two: the second-order capacitance data C2 is input into the neural network model, and the constraint condition is added, and third-order capacitance data C3 is obtained;

[0043] Step4, image reconstruction: the third-order capacitance data C3 is input into the finite element analysis model, and the image in the to-be-measured region is obtained;

[0044] Step5, evaluation and optimization: the reconstructed image is evaluated, if the imaging is not clear, the data processing two and the image reconstruction process are repeated until the imaging quality meets the requirements.

[0045] In the embodiment of the application, the initial capacitance data C1 is processed by principal component analysis method to reduce dimension, which can effectively reduce the dimension of input data, remove redundant information and only retain the most important feature set in the initial capacitance data C1. Considering that the principal component analysis method is linear dimension reduction, there are some nonlinear original data in the initial capacitance data C1, therefore, only using the principal component analysis method may not be able to capture the important features of the initial capacitance data C1, therefore, the second-order capacitance data C2 after dimension reduction is further processed, that is, processed by a neural network model. The neural network model can capture, fit and other operations on nonlinear features under the action of the constraint condition, and output third-order capacitance data C3. Therefore, a suitable neural network model can be selected or designed to make the neural network model establish a more accurate mapping relationship between the second-order capacitance data C2 and the third-order capacitance data C3, so that the neural network model can capture or improve the data missed by the principal component analysis method. In this way, the second-order capacitance data C2 after principal component analysis reduces the computational complexity of the neural network model, speeds up the training speed, reduces the risk of overfitting, and helps the neural network model to better learn the key features in the data, thereby establishing a more concise and accurate neural network model. And the neural network model can perform secondary processing on the second-order capacitance data C2 to finally output accurate third-order capacitance data C3. The whole process consumes less time, the calculation amount is moderate, and the data accuracy can be guaranteed. Finally, the image reconstruction is performed by using the finite element analysis model, and the third-order capacitance data C3 has been processed by dimension reduction and neural model improvement, which can maintain a certain accuracy and simplicity. After the image reconstruction is performed by using the finite element analysis model, the time consumption is shorter and the result is more accurate.

[0046] In an optional embodiment of the application, for Step1, data acquisition.

[0047] Specifically, it includes: applying an excitation signal, applying an excitation signal through one electrode of the capacitive sensor, the excitation signal can be an alternating signal or a pulse signal, which is used to excite the charge distribution in the to-be-measured region; measuring the capacitance value, measuring the capacitance value through the other electrode of the capacitive sensor, the capacitance value can be measured using a capacitance measurement circuit, connecting the capacitive sensor to a measurement device, in one example, the measurement device can be a computer, a single-chip microcomputer, or the like terminal; collecting multiple sets of data, repeating the above two steps, collecting multiple sets of capacitance data and recording to form initial capacitance data C1.

[0048] In one of the optional embodiments of the present application, for Step2, data processing one.

[0049] Specifically, it includes: standardizing the initial capacitance data C1, so that the mean value of each feature is zero and the variance is one, which is to ensure that the importance of each feature is relatively balanced, and to avoid that some features have too much influence on the principal component analysis result; calculating the covariance matrix of the standardized capacitance data C1, the covariance matrix is used to indicate the correlation between different features; performing eigenvalue decomposition on the covariance matrix to obtain eigenvalues and corresponding eigenvectors, wherein the eigenvalues represent the importance of each feature, and the eigenvectors represent the projection of each feature in the principal component direction; sorting the eigenvalues in descending order, and selecting the eigenvectors corresponding to the first k eigenvalues as the principal components, in this step, the principal components are selected, and k is the dimension after dimension reduction, which can be selected according to needs; projecting the standardized capacitance data C1 onto the selected principal components, i.e., multiplying the initial capacitance data C1 by the selected principal component eigenvectors to obtain the reduced second-order capacitance data C2.

[0050] For Step3, data processing two.

[0051] Specifically, it includes using the trained neural network model, inputting the second-order capacitance data C2 into the model, and obtaining the predicted third-order capacitance data C3 through the forward propagation process of the model. The constraint conditions include physical law constraints: according to the physical law of the problem, the constraint condition can be established on the basis of the basic physical principle. For example, in the capacitance problem mentioned in the present application, the constraint condition satisfies a specific mathematical formula or relationship between the capacitance value and the dielectric constant, or the number of electrodes, the current intensity, etc. For example, the capacitance value C satisfies the following formula between the dielectric constant, the electrode area A and the electrode distance d: C = ε r *A / d.

[0052] In other optional examples, the constraint conditions can also include prior knowledge constraints: according to known prior knowledge or experience, the constraint conditions can be added to the model. For example, according to existing data or the knowledge of domain experts, the output of the constraint model is within a certain range or satisfies a certain distribution. Data constraints: according to the characteristics and limitations of the data, constraint conditions can be added. For example, the range, distribution or specific data properties of the input data can be constrained. Parameter constraints: constraints are placed on the parameters of the model to limit their value range or satisfy certain conditions. For example, the weight matrix of the model can be constrained to have sparsity or satisfy certain regularization conditions.

[0053] It should be noted that between data processing one and data processing two, there is also a step Step0: establishing a neural network model.

[0054] Specifically, it includes: Step01, solving the positive problem: according to the initial capacitance data C1 collected and the characteristic parameters of the collection system, the theoretical capacitance value distribution is calculated, denoted as the target capacitance data C0; Step02, setting a neural network model suitable for processing capacitance data, and selecting a suitable network structure and layer number, and adding constraint conditions; Step03, taking the second-order capacitance data C2 as the input data and the target capacitance data C0 as the output data, training the neural network model through the second-order capacitance data C2 and the target capacitance data C0, so that the neural network model learns the mapping relationship between the input and output data.

[0055] In the embodiments of the present application, the target capacitance data C0 is obtained by solving the forward problem. The target capacitance data C0 is a set of data that can most accurately reflect the image of the to-be-measured region in theory, and is obtained by a series of rigorous formula derivation and calculation, which is very mature in the prior art and will not be described here. The traditional process of establishing a neural network model is as follows: input the initial capacitance data C1 and the target capacitance data C0, and train the neural network multiple times to learn the mapping relationship between C1 and C0, and then input the corresponding initial capacitance data C1 to obtain the capacitance data used for image reconstruction. However, this has obvious problems. That is, the initial capacitance data C1 has a large amount of data interference, and the data volume itself is large, which greatly increases the learning time and load degree of the neural network model, and also greatly interferes with the mapping accuracy of the neural network model, which greatly hinders the instantaneity of image reconstruction. In the present application, the initial capacitance data C1 is processed by the principal component analysis method, and the processed second-order capacitance data C2 is input during the training of the neural network. Then, the neural network model only needs to learn the relationship between C2 and C0. In this way, the training process of the neural network model is greatly reduced, and since the guide capacitance data is also the target capacitance data C0, the third-order capacitance data C3 used for the image reconstruction process can be obtained when C2 is input subsequently, and the third-order capacitance data C3 is relatively close to the target capacitance data C0.

[0056] As one of the optional embodiments of the present application, the steps of solving the forward problem include: establishing a finite element model: dividing the target object into discrete small cells, each cell having a certain electrical conductivity; constructing a forward problem equation: constructing a forward problem equation according to the initial capacitance data C1 and the finite element model, the forward problem equation being used to indicate the relationship between the initial capacitance data C1 and the electrical conductivity of the to-be-measured region; solving the forward problem: obtaining the electrical conductivity distribution of the to-be-measured region by solving the forward problem equation, denoted as the target capacitance data C0.

[0057] Specifically.

[0058] Establishing a finite element model: First, a finite element model of the object is established, which describes the geometric shape and capacitance characteristics of the object. The object region is divided into many small units, such as triangles or quadrilaterals (for two-dimensional cases) or tetrahedrons or hexahedrons (for three-dimensional cases), and appropriate basis functions are selected to describe the capacitance distribution within each unit.

[0059] Determining boundary conditions: According to the actual situation, the boundary conditions are determined. The boundary conditions include the position and shape of the capacitance probe, the potential or current input of the capacitance probe, etc.

[0060] Establishing a forward problem model: according to the finite element model and the boundary conditions, a forward problem model is established. The forward problem model describes the relationship between the measured capacitance value of the sampling probe and the object capacitance distribution. By solving the forward problem model, the theoretical capacitance value distribution can be obtained.

[0061] The relationship between the capacitance value C and the dielectric constant ε distribution can be expressed as follows:

[0062]

[0063] In the formula: ε(x, y) is the dielectric constant distribution function of the measured field, is the potential distribution function, V is the potential difference between the source electrode and the detection electrode, and Γ is the electrode surface. Solving the dielectric constant matrix G distribution from the capacitance value matrix C and the sensitive field matrix S is the image reconstruction of ECT.

[0064] Meshing: according to the geometry of the object and the finite element model, the object region is meshed. Meshing divides the object region into many small units for numerical calculation.

[0065] Solving the forward problem model: using numerical calculation methods such as finite element method to solve the forward problem model. By solving the model equation, the capacitance value distribution is calculated.

[0066] As one of the optional embodiments of the present application, the step of evaluating and optimizing further includes: if the results of multiple image reconstructions do not meet expectations, replacing different types of neural network models and repeating the step of establishing a neural network model, wherein the neural network model includes a convolutional neural network, a recurrent neural network, and a feedforward neural network.

[0067] In the embodiments of the present application, since there are many types of neural network models and the adaptability is also different, if the results of multiple image reconstructions do not meet expectations, it proves that the neural network model may have the problem of poor adaptability. In the process of evaluating and optimizing, the neural network model is changed to find the most suitable neural network model, thereby improving the accuracy of subsequent data processing.

[0068] As one of the optional embodiments of the present application, the step of evaluating and optimizing further includes: if the results of multiple image reconstructions do not meet expectations, changing the combination of constraint conditions for repeated training, wherein the constraint conditions include physical constraints, spatial constraints, prior knowledge constraints, etc.

[0069] In the embodiments of the present application, by trying multiple combinations of constraint conditions, the most suitable constraint condition for the current detection process can be found in the neural network training process, so as to remove the noise in the capacitance data to the greatest extent, improve the smoothness of the data, and further improve the spatial resolution of the capacitance image, so that the imaging result is more accurate and detailed.

[0070] As one of the optional embodiments of the present application, before the data processing step, a data preprocessing step is further included: processing the abnormal values and noise in the initial capacitance data C1.

[0071] Specifically, noise removal: noise is a common interference factor in capacitance measurement data, which may come from electronic devices, environmental interference, etc. In order to improve the data quality, filtering technology can be used for noise removal processing. Common filtering methods include mean filtering, median filtering, Gaussian filtering, etc. According to the actual situation, select the appropriate filtering method. Correction: in ECT technology, the capacitance measurement data may be affected by factors such as electrode spacing and electrode shape, resulting in deviation in the measurement results. In order to eliminate these deviations, correction operation is needed. Correction can be carried out through experimental measurement and numerical simulation, etc. According to the actual situation, select the appropriate correction method. Normalization: in order to make different measurement data comparable, the capacitance measurement data can be normalized. Common normalization methods include maximum and minimum value normalization, mean and variance normalization, etc. Normalization can map data to a specific range, so that the data has similar scale and distribution characteristics.

[0072] In the principal component analysis method in the embodiment of the present application, there may be a case of insufficient data capture, which is due to the collection of abnormal values or noise in the initial capacitance data C1 due to circuit noise, capacitance sensor installation position and precision control, etc. Since the principal component analysis method is based on the calculation of the covariance matrix or correlation matrix of the data, the existence of abnormal values or noise may cause the calculation result of the covariance matrix to deviate from the true situation, thereby affecting the extraction and interpretation of the principal components. Therefore, data preprocessing in advance can reduce the influence on the result of the principal component analysis process and ensure the accuracy of the result.

[0073] As one of the optional embodiments of the present application, the characteristic parameters of the acquisition system include: the geometric shape of the capacitance sensor, the electrode spacing, the electrode size, etc.

[0074] As one of the optional embodiments of the present application, after the step of evaluating and optimizing, it further includes: if the image shows that the particle diameter D is greater than or equal to 0.5 cm, or the image shows that the solubility is continuously high within each monitoring period T minutes, an alarm information is sent, wherein 15≤T≤25.

[0075] In another aspect, the application provides a lubricating oil detection system, comprising: a plurality of capacitive sensors arranged in a ring shape on the outer wall of a pipeline; a data acquisition module, comprising a driving unit and a transmission unit, the driving unit being configured to apply an alternating voltage to stimulate a region to be measured; and a terminal configured to analyze and process the acquired capacitive data and reconstruct an image; wherein the transmission unit is electrically connected to the capacitive sensors and configured to convert the current signals into digital signals and transmit the digital signals to the terminal.

[0076] The above merely describes some embodiments of the application, and is not intended to limit the application in any form. The scope of protection of the embodiments of the application is not limited thereto. Any simple modification, equivalent change and modification that can be easily thought of by those skilled in the art within the technical scope disclosed by the embodiments of the application shall be encompassed within the scope of protection of the embodiments of the application.

Claims

1. A method for detecting lubricating oil wear particles, characterized in that: include: Data acquisition: Capacitive sensors are placed on the outer wall of the pipeline to obtain the initial capacitance data C1 of the area to be measured; Data processing 1: performing dimensionality reduction processing on the initial capacitance data C1 by principal component analysis to obtain second-order capacitance data C2; Data processing 2: Input the second-order capacitance data C2 into the neural network model, add constraints, and obtain the third-order capacitance data C3; Image reconstruction: Input the third-order capacitance data C3 into the finite element analysis model to obtain the image of the area to be measured; Evaluation and optimization: Evaluate the reconstructed image. If the image is not clear, repeat the data processing and image reconstruction process until the image quality meets the requirements.

2. The lubricating oil wear particle detection method according to claim 1, characterized in that: Between data processing 1 and data processing 2, there is also the step of establishing a neural network model: Solution to the forward problem: Based on the collected initial capacitance data C1 and the characteristic parameters of the acquisition system, calculate the theoretical capacitance value distribution and record it as the target capacitance data C0; Set up a neural network model suitable for processing capacitance data, select the appropriate network structure and number of layers, and add constraints; The second-order capacitance data C2 is used as input data, and the target capacitance data C0 is used as output data. The neural network model is trained using the second-order capacitance data C2 and the target capacitance data C0 so that the neural network model learns the mapping relationship between the input and output data.

3. The lubricating oil wear particle detection method according to claim 2, characterized in that: The steps of solving the positive problem include: Establish a finite element model: divide the target object into discrete small cells, each of which has a certain conductivity; Constructing a forward problem equation: constructing a forward problem equation based on the initial capacitance data C1 and the finite element model, wherein the forward problem equation is used to indicate the relationship between the initial capacitance data C1 and the conductivity of the area to be measured; Direct problem solution: By solving the direct problem equation, the conductivity distribution of the area to be measured is obtained and recorded as the target capacitance data C0.

4. The lubricating oil wear particle detection method according to claim 2, wherein: The evaluation and optimization step also includes: if the results of multiple image reconstructions do not meet expectations, different types of neural network models are replaced, and the step of establishing the neural network model is repeated, wherein the neural network model includes a convolutional neural network, a recurrent neural network, and a feedforward neural network.

5. The lubricating oil wear particle detection method according to claim 4, characterized in that: The evaluation and optimization step also includes: if the results of multiple image reconstructions do not meet expectations, changing the combination of constraints and performing repeated training, wherein the constraints include physical constraints, spatial constraints, prior knowledge constraints, etc.

6. The lubricating oil wear particle detection method according to claim 2 or 5, characterized in that: Before the data processing step, a data preprocessing step is also included: processing abnormal values ​​and noise in the initial capacitance data C1.

7. The lubricating oil wear particle detection method according to claim 1, wherein: The step of performing dimensionality reduction processing on the initial capacitance data C1 by using a principal component analysis method to obtain second-order capacitance data C2 includes: The initial capacitance data C1 is normalized so that the mean of each feature is 0 and the variance is 1; Calculate the covariance matrix of the standardized capacitance data C1; Perform eigenvalue decomposition on the covariance matrix to obtain the eigenvalues ​​and corresponding eigenvectors; Sort the eigenvalues ​​in descending order, and select the eigenvectors corresponding to the first k eigenvalues ​​as the principal components; The normalized capacitance data C1 is projected onto the selected principal component to obtain the second-order capacitance data C2 after dimensionality reduction.

8. The lubricating oil wear particle detection method according to claim 2, characterized in that: The characteristic parameters of the acquisition system include: the geometric shape of the capacitance sensor, the electrode spacing, the electrode size, etc.

9. The lubricating oil wear particle detection method according to claim 1, wherein: After the evaluation and optimization step, the method further includes: issuing a warning message if the image shows that the particle diameter D is greater than or equal to 0.5 cm, or the image shows that the solubility is continuously high within T minutes of each monitoring cycle, wherein 15≤T≤25.

10. The lubricating oil wear particle detection method according to claim 1, characterized in that: It uses a lubricating oil detection system, which includes: Capacitive sensors are provided in multiple numbers and are evenly arranged in a ring shape on the outer wall of the pipeline; The data acquisition module includes a driving unit and a transmission unit, wherein the driving unit is used to apply an alternating voltage to stimulate the area to be measured; A terminal is used to analyze and process the collected capacitance data and reconstruct an image; The transmission unit is electrically connected to the capacitive sensor to convert the current signal into a digital signal and transmit the digital signal to the terminal.

Images